Embed Size (px)
Citation preview
Ecological Informatics 30 (2015) 40–48
Contents lists available at ScienceDirect
Ecological Informatics
j ourna l homepage: www.e lsev ie r .com/ locate /eco l in f
Using custom scientific workflow software and GIS to inform protectedarea climate adaptation planning in the Greater Yellowstone Ecosystem
N.B. Piekielek a,⁎, A.J. Hansen b, T. Chang b
a University Libraries, 208L Paterno Library, The Pennsylvania State University, University Park, PA 16802, United StatesbDepartment of Ecology, Lewis Hall, Montana State University, Bozeman, MT 59717, United States
⁎ Corresponding author. Tel.: +1 814 865 3703.E-mail addresses: [email protected] (N.B. Piekielek), h
(A.J. Hansen), [email protected] (T. Chang).
http://dx.doi.org/10.1016/j.ecoinf.2015.08.0101574-9541/© 2015 Elsevier B.V. All rights reserved.
a b s t r a c t
a r t i c l e i n f oArticle history:Received 17 April 2015Received in revised form 18 August 2015Accepted 21 August 2015Available online 10 September 2015
Keywords:Climate changeGreater Yellowstone EcosystemSpecies distribution modelAdaptive management
Anticipating the ecological effects of climate change to inform natural resource climate adaptation planning rep-resents one of the primary challenges of contemporary conservation science. Species distribution models havebecome a widely used tool to generate first-pass estimates of climate change impacts to species probabilitiesof occurrence. There are a number of technical challenges to constructing species distribution models that canbe alleviated by the use of scientific workflow software. These challenges include data integration, visualizationof modeled predictor–response relationships, and ensuring that models are reproducible and transferable in anadaptive natural resource management framework. We used freely available software called VisTrails Softwarefor Assisted Habitat Modeling (VisTrails:SAHM) along with a novel ecohydrological predictor dataset and the lat-est CoupledModel Intercomparison Project 5 future climate projections to construct species distribution modelsfor eight forest and shrub species in the Greater Yellowstone Ecosystem in the Northern Rocky Mountains USA.The species considered included multiple species of sagebrush and juniper, Pinus flexilis, Pinus contorta,Pseudotsuga menziesii, Populus tremuloides, Abies lasciocarpa, Picea engelmannii, and Pinus albicaulis. Current andfuture species probabilities of occurrence were mapped in a GIS by land ownership category to assess the feasi-bility of undertaking present and future management action. Results suggested that decreasing spring snowpackand increasing late-season soil moisture deficit will lead to deteriorating habitat area for mountain forest speciesand expansion of habitat area for sagebrush and juniper communities. Results were consistent across nine globalclimate models and two representative concentration pathway scenarios. For most forest species their projectedfuture distributions moved up in elevation from general federal to federally restricted lands where active man-agement is currently prohibited by agency policy. Though not yet fully mature, custom scientific workflow soft-ware shows considerable promise to ease many of the technical challenges inherent in modeling the potentialecological impacts of climate change to support climate adaptation planning.
© 2015 Elsevier B.V. All rights reserved.
1. Introduction
Anticipating the ecological effects of climate change to inform natu-ral resource climate adaptation planning represents one of the primarychallenges of contemporary conservation science (Rehfeldt et al., 2012).Addressing this challenge will require advances in basic ecological un-derstanding, the capabilities of scientific modeling tools and efficientuse of limited resources. Large scale and long-term experimentation(e.g., Harte et al., 2006) has contributed to addressing thefirst challenge,whereas recent development of scientific workflow software showspromise of contributing to the latter two challenges. The present studydemonstrated how use of a custom species distribution modeling
software helped to inform climate adaptation planning in a protectedecosystem of high conservation interest.
Species distributionmodels (SDMs) have become a common tool forassessing the potential impact of climate change on species (Araújo andGuisan, 2006). SDMs are a correlative approach that identifies species'climate tolerances with a goal of projecting changes in future probabil-ities of occurrence under alternative climate change scenarios. SDMs arebased on niche theory (Franklin, 2009) andmake a number of necessaryassumptions including: that present day species distributions and theirrelationship to climate adequately captures the actual tolerances of spe-cies that should be mapped in future scenarios; that predictors capturethe direct effects of climate on species distributions that will generalizewell to future conditions; and that other factors that influence distribu-tion dynamics (e.g., dispersal, and competition) play comparativelyminor roles so that mapping probability of occurrence is a reasonablefirst step to understanding the ecological consequences of climatechange (Anderson, 2013). There are a number of technical challenges
41N.B. Piekielek et al. / Ecological Informatics 30 (2015) 40–48
in constructing species distribution models including that they oftenutilizemany large spatial datasets that are time-consuming to integrate,they rely on sophisticated statistical methods that can obscure the rela-tionships between predictors and species response that are representedby models, and they often require iterative learning and analysis thatcan be difficult to document so that models are repeatable and transfer-able. As such, there is increasing interest in customizing scientificworkflow software to ease some of the technical burdens of species dis-tributionmodeling to support natural resource climate adaptation plan-ning (Morisette et al., 2013).
The results of recent national and continental scale SDMstudies pro-vide different insights on the potential future impact of climate changeon vegetation in the Northern Rocky Mountains in general, and GreaterYellowstone Ecosystem (GYE) in particular. In a study of the continentalU.S., Rehfeldt et al. (2012) projected a sharp decline in climate suitabil-ity for the Rocky Mountain Subalpine Conifer Forest biome and an ex-pansion of Great Basin Scrub. Coops and Waring (2010) applied theirPhysiological Principles Predicting Growth (3-PG) model to the PacificNorthwest and suggested that climate change may lead to habitat ex-pansion for warm mesic forest species like Pseudotsuga menziesii. Also,Bell et al. (2014) reported decreasing climate suitability for threeupper treeline forest species in the dry western U.S. and increasing suit-ability for one xeric woodland species. In studies focused specifically onthe GYE, Schrag et al. (2008) projected decreases in habitat suitabilityfor three upper treeline species under an increased temperature andprecipitation scenario, but slight increases in suitable area under an in-creased precipitation scenario. Finally, Chang et al. (2014) projecteddramatic declines in climate suitability for one upper treeline forest spe-cies. Differences in resultsmost likely came frommodel formulation, thechallenge of capturing the direct effects of climate on vegetation, andchanging future climate projections.
Recent advances in the development of climate-derived predictorsthat are thought to better represent the direct effects of climate on vege-tation and the availability of the most recent global climate models(GCMs) now provide the opportunity to revisit the expected impact ofclimate change on GYE vegetation. In particular, dynamic water-balancemodels (e.g., Lutz et al., 2010) use commonly available climate andother data to derive SDM predictors that are thought to better representthe direct effects of climate on vegetation than do summaries of temper-ature and precipitation alone (Stephenson, 1998). The CoupledModel In-tercomparison Project (CMIP) Phase 5 GCMs have made improvements
Fig. 1. The Greater Yellowstone Ecosystem study area by land management type. The black ouproduction online only].
over Phase 3models by better representing radiative forcings and includ-ing more physical atmospheric processes (Knutti and Sedláček, 2012).
To address the uncertainties apparent in prior studies and demon-strate the use of scientific workflow software for climate adaptationstudies, we developed the following study objectives:
1. Evaluate the potential impact of climate change on dominant treespecies and sagebrush in four elevation zones of the GYE usingSDMs under multiple CMIP Phase 5 climate change scenarios.
2. Interpret SDM results relative to projected changes in the mean ele-vation and management jurisdiction of projected future speciesdistributions.
3. Evaluate the use of custom scientific workflow software to supportnatural resource climate adaptation planning.
2. Materials and methods
2.1. Study area
The Greater Yellowstone Ecosystem covers portions of Montana,Wyoming, and Idaho in the Northern Rocky Mountains USA (Fig. 1). Itis commonly regarded as one of themost ecologically intact ecosystemsin the continental U.S. (Schrag et al., 2008). Over half of the GYE isprotected for conservation and related purposes, mostly by the U.S. Na-tional Park Service (NPS) and U.S. Forest Service (USFS). There is a longhistory of active management on general USFS lands whereas manage-ment policies often preclude its use on NPS lands as well as within des-ignated wilderness, proposed wilderness, and roadless areas (referredto collectively herein as federally restricted lands). Surrounding the Yel-lowstone plateau (the middle elevations on which NPS lands are cen-tered) are mountains (up to 4000 m), beyond which lower elevationvalleys and steppe predominate and ownership grades to almost entire-ly private where coordinated natural resource management is unlikely.The GYE supports numerous iconicwilderness species that are the focusof national conservation policy and debate. The primary future threatsto biodiversity conservation in the GYE include future land use(Piekielek and Hansen, 2012), and climate change (Hansen et al., 2014).
The region experiences a cold continental climate with more warmand dry conditions at lower elevations and more cool and wet condi-tions at higher elevations and across much of the Yellowstone plateau.Soils of the plateau tend to be of rhyolitic volcanic origin and are
tline shows the National Park Service boundary for reference [2-column graphic, color-re-
Table 1Environmental predictors used to model probability of occurrence for eight species in theGreater Yellowstone Ecosystem under current and projected future climate changescenarios.
Predictor Abbreviation Data source
August soil moisturedeficit
deft9 PRISM/CMIP5 temperature,precipitation, soils and topography asinput to water-balance model(Lutz et al., 2010)
April snowpack pack4 PRISM/CMIP5 temperature,precipitation, soils and topography asinput to water-balance model(Lutz et al., 2010)
June soil moisture soilm6 PRISM/CMIP5 temperature,precipitation, soils and topography asinput to water-balance model(Lutz et al., 2010)
Fraction of sand intop 100 cm of soil
sandfract CONUS-SOIL (Miller and White, 1998)
Fraction of rock byvolume in the top100 cm of soil
rckvol CONUS-SOIL (Miller and White, 1998)
Direct incomingsolar radiation
srad USGS 30-meter digital elevationmodel as input to ArcGIS SpatialAnalysis tools (ESRI, 2009)
Topographicwetness index
twi USGS 30-meter digital elevationmodel as input to ArcGIS SpatialAnalysis tools (ESRI, 2009)
42 N.B. Piekielek et al. / Ecological Informatics 30 (2015) 40–48
sandy with poor water-holding capacity and little nutrient availabilityto support plant growth (Despain, 1990). Soils throughout the sur-rounding mountains are often of andesitic volcanic origin and tend tobe dominated by clays, have higher water-holding capacities and arecomparatively nutrient rich. The climate of the GYE has already begunto change relative to periods prior to 1980, with spring and summertemperatures approximately b1 °C warmer and reduced spring andsummer snowpack (Westerling et al., 2011). The future climate of theGYE is expected to be warmer and drier than present (Chang et al.,2014). More complete descriptions of the climate, major vegetationtypes, and soils and topography of the GYE are provided by Brownet al. (2006) and Despain (1990).
2.2. Focal species
The present study focused on the dominant vegetation species of fourelevation zones being sagebrush communities (up to ~1900 m), lowertreeline forests (up to ~2200 m), montane forests (up to ~2500 m),and upper treeline forests (up to ~3000 m). Sagebrush tends to occupyareas just downslope of treeline as a result of its ability to persist onwell-drained soils that have a soilwater deficit late in the growing season(Despain, 1990; Schlaepfer et al., 2012). Multiple species of sagebrushwere all grouped into a single sagebrush community category for thepresent study. These species included: big sagebrush (Artemesiatridentata Nutt.), mountain big sagebrush (A. tridentata Nutt. spp.vaseyana), threetip sagebrush (A. tridentata Rydb. ssp. tripartita), Wyo-ming big sagebrush (A. tridentata Nutt. spp. wyomingensis), and basinbig sagebrush (A. tridentata Nutt. spp. tridentata) (http://plants.usda.gov/). Lower treeline species included limber pine (Pinus flexilis) that isoften found on steep, dry, rocky soils in toe slope settings (Means,2011), and three species of juniper (Juniperus scopulorum, Juniperusoccidentalis, and Juniperus osteosperma) that are generally well-adaptedto hot and dry growing conditions (Despain, 1990). All juniper specieswere grouped into a single juniper community category for the presentstudy. Montane forest species included lodgepole pine (Pinus contorta),Douglas fir (P. menziesii), and quaking aspen (Populus tremuloides). Mon-tane forests included a mix of species that require deep nutrient richsoils, longer growing seasons, and adequate moisture (Douglas fir andaspen), and those that are comparatively cold and drought hardy andcan tolerate lower soil nutrients (lodgepole pine). Soil properties arethought to determine themembership ofmontane forests across the Yel-lowstone plateau with lodgepole pine present on the nutrient poorsandy soils of rhyolitic parent material to the exclusion of Douglasfir and at higher elevations Englemann spruce and subalpine fir(Brown et al., 2006; Coops and Waring, 2010; Littell et al., 2008;McLane et al., 2011). The highest elevation forests occur below anupper treeline that is the result of cold temperature limitations andlong durations of annual snow cover as well as shallow soils and/oran absence of soil substrate where there are rock outcroppings (Schraget al., 2008). Typical upper treeline forest species include subalpine fir(Abies lasciocarpa), Englemann spruce (Picea engelmannii), and white-bark pine (Pinus albicaulis).Whitebark pinewas not included in the pres-ent study because it was considered in a companion study (Chang et al.,2014).
2.3. Response data
Wederived presence and absence data for each species from theU.S.Forest Service Forest Inventory and Analysis (FIA) program database(http://apps.fs.fed.us/fiadb-downloads/datamart.html). The FIA pro-gram employs a regular gridded sampling design with one field plotfor approximately every 2500 ha. Individual tree species and measure-ments are recorded during field visits for trees that are larger than12.7 cm diameter at breast height. Information on understory specieslike sagebrush is recorded for a subset of FIA plots (see O'Connell,et al., 2012 for a description of FIA sampling methods).
The precise locations of FIA plots are protected. Imprecise plot loca-tion can affect the quality of SDMs by associating species with habitatconditions that do not actually support them (Gibson et al., 2013). Astudy that examined the effect of building SDMs with precise versus im-precise FIA plot location data found little effect, with a few exceptions(Gibson et al., 2013). One exception was for a species considered in thepresent study (J. occidentalis). As such, we undertook two additionaldata preparation steps. First, we compared the fieldmeasured plot eleva-tion reported in the FIA database to the elevation of the plot's impreciselocation by intersecting it with a 30-m digital elevation model and elim-inated plots that exhibited more than a 333-m discrepancy. This wassimilar to what has been done by other studies in the region (Coopsand Waring, 2010). Second, we screened our presence and absencedata with a modeled data product produced with precise FIA plot loca-tions (Wilson et al., 2012). Finally, to maintain sample independence,which would have been violated by associating multiple presence/ab-sence observations with the exact same environmental conditions (i.e.,within the same predictor grid cell), we retained only one FIA plot foreach predictor grid cell. Where there was a presence and absence obser-vation within a single predictor cell the presence observation wasretained in order to identify the broadest possible environmental toler-ances for each species.
We used observations from 2489 FIA plots to build forest SDMs and1374 plots to build sagebrush SDMs. We chose to limit the area of anal-ysis to the GYE rather than entire species ranges because more detailedpredictor data were available across this study domain, some speciesexhibit distinct genotypes that are adapted to local conditions and wewanted to produce results for a management relevant study area (i.e.,the area crosswhichmanagement action could conceivably have an im-pact). The choice of a limited study areamay have identified species tol-erances that were narrower than had we examined their entire range.Narrower than actual tolerances would lead to inflated potential impact(i.e., change in distribution area) when projected to future conditions.Given these limitations and recognizing the short-comings of a correla-tive study, we focus our discussion on potential impacts of species andcommunities relative to one and other rather than on the specific mag-nitude of projected range expansion and contraction.
43N.B. Piekielek et al. / Ecological Informatics 30 (2015) 40–48
2.4. Predictor data
To expand on prior SDM efforts and test a suite of predictors thatshould better represent the direct effects of climate on vegetation wederived predictor data related to monthly water-balance, soils and to-pography at a 30 arc sec spatial resolution (Table 1). Estimates ofmonthly temperature and precipitation that were inputs to a water-balance model (sensu Lutz et al., 2010) came from the griddedParameter-elevation Regressions on Independent Slopes Model dataset(PRISM Climate Group, Oregon State University, http://prism.oregonstate.edu). PRISM data represent interpolations of weather sta-tion observations based on location, elevation and other predictors ofspatial variation in climate at regional scales (Daly et al., 2008). Thewater-balance model was a dynamic “bucket model” that added soilwater or snowpack on a monthly time step and consumed availablesoil moisture (or released it as runoff or groundwater recharge) usingapproximated rates of evapotranspiration based on temperature withadjustments for latitude, slope, and aspect. A dynamic water-balancemodel that carried snow and soil moisture forward from one monthto the next was thought to be an important improvement over temper-ature and precipitation predictors in the GYE because snow dynamicsplay a large role in controlling seasonal soil moisture availability.Water-balance predictors were also found to be important in a compan-ion study (Chang et al., 2014). Water-balance predictors were summa-rized as 30-year monthly averages for the time-period 1951–1980.
To represent spatial variation in soil properties we used CONUS-SOILdata from Miller and White (1998). These data describe soil physicaland chemical properties that likely limit the distributions of some spe-cies and influence the outcomes of local biotic interactions such as com-petition (Despain, 1990). Soil predictors included the fraction byvolume of rock fragments (larger than 2 mm unattached) and sand inthe top 100 cm of surface soil. From USGS 30 meter digital elevationmodels (Gesch et al., 2009) and ArcGIS 9.3 Spatial Analyst tools (ESRI,2009) we derived a topographic wetness index, and direct solar radia-tion predictors. Soils and topography were included in the presentstudy because they are of known local importance to forest communitymembership (Despain, 1990; Schrag et al., 2008), however, these vari-ables have not been included in prior national and continental scalestudies.
2.5. Future climate data
To project future species distributionswe chose nine GCMs based ontheir ability to match 20th century weather observations in the region(Rupp et al., 2013) (Table 2). For each GCM, we also used a high (8.5)and low (4.5) representative concentration pathway (RCP) that refersto the amount of anthropogenic climate forcing in watts per squaremeter. The RCP 8.5 scenario is consistent with increases in atmosphericgreenhouse gases at rates similar to present, whereas the RCP 4.5
Table 2List of nine global climate models chosen based on their mean error relative to observed20th century climate in the region as evaluated by Rupp et al. (2013).
Model name Developer
CESM1-CAM5 Community Earth System Model ContributorsCESM1-BGC Community Earth System Model ContributorsCCSM4 Community Earth System Model ContributorsCNRM-CM5 Centre National de Recherches Météorologiques/Centre Européen
de Recherche et Formation Avancée en Calcul ScientifiqueHadGEM2-AO National Institute of Meteorological Research/Korea
Meteorological AdministrationHadGEM2-ES Met Office Hadley Centre (additional HadGEM2-ES realizations
contributed by Instituto Nacional de Pesquisas Espaciais)HadGEM2-CC Met Office Hadley Centre (additional HadGEM2-ES realizations
contributed by Instituto Nacional de Pesquisas Espaciais)CMCC-CM Centro Euro-Mediterraneo per I Cambiamenti ClimaticiCanESM2 Canadian Centre for Climate Modelling and Analysis
scenario would require aggressive and coordinated reductions in globalgreenhouse gas emissions (Moss et al., 2010). Monthly projected futuretemperature and precipitation data were downloaded from the NASANCCS THREDDS website (https://cds.nccs.nasa.gov/nex/) where down-scaling was performed according to Thrasher et al. (2013). Future hab-itat was considered to be where SDMs run with at least five of nine(i.e., a simple majority) GCM inputs agreed on species presence. Wealso considered other threshold levels of agreement varying from oneGCM to all nine GCMs and reported the variation in future suitable hab-itat area.
2.6. Modeling methods
We used a freely available open source scientific workflow softwareVisTrails (Freire and Silva, 2012), that has been customized for use inSDM studies using a set of add-on tools that comprise the Software forAssisted Habitat Modeling (SAHM) (Morisette et al., 2013; https://www.fort.usgs.gov/products/23403). VISTRAILS:SAHM provides userswith tools for data preprocessing including automated reprojection, ag-gregation, resampling and subsetting of predictor data to match a tem-plate layer. VISTRAILS:SAHM users can select one of five differentstatistical SDM methods. Upon running a model, the VISTRAILS:SAHMtool produces model diagnostics including the generation of responsecurves that graphically represent the relationships between predictorsand response that are represented in models. To save a record of thedata used and workflow provenance, VISTRAILS:SAHM saves the inputparameters and outputs of each model run so that a user can easily rec-reate any exploratory step and transfer final models to other interestedresearchers or natural resource managers.
The goal of model construction was to identify a common set of pre-dictors that performed well for all species in terms of common SDM di-agnostics and matching our understanding of GYE ecology. Modelconstruction was performed as follows. We first conducted a literaturereview of the environmental factors that have been hypothesized tolimit each species.We next searched for the best available data to repre-sent these predictors. Many predictors were found to be collinear. Col-linearity can increase model coefficient standard errors and lead tolarge model prediction errors. We therefore, selected only predictorsthat were less strongly correlated (considered b 0.7 of the maximumPearson, Spearman, and Kendall's correlation coefficients). After exten-sive exploratory analyses of different statistical methods and predictorsets, we chose to use multivariate adaptive regression splines (MARS)(Leathwick et al., 2006) to build SDMs.MARSmethods fit non-linear re-lationships between environmental predictors and species presencewith piecewise basis functions across multiple breakpoints (i.e., knotpoints) in an intentionally overfit forward stepwise manner and thenprunes these based on their contribution to the model. Final binary oc-currencemapswere produced using a probability of presence thresholdso that sensitivity (i.e., true positive rate) and specificity (i.e., true neg-ative rate) were approximately equal.
To evaluate competing models, we produced three typical SDM di-agnostics through a 33%, 10-fold cross-validation. In cross-validation,models were trained on an approximately random selection of 67% ofthe dataset and model performance evaluated against the withheld33% (Leathwick et al., 2006). This was repeated ten times andmodel di-agnostics were reported as the average of all ten steps. We also exam-ined response curves, which graphically represent the relationshipbetween model predictors and fitted values (i.e., probabilities of pres-ence). Response curves that are truncated, or suffer from anomalies atsampled range edges (e.g., a sudden change in the relationship fromnegative to positive), can lead to results that are inconsistent with eco-logical understanding (Anderson, 2013). Models that produced re-sponse curves that were inconsistent with ecological understanding orthat produced anomalies were thrown out, even if they exhibitedgood model fit based on other model diagnostics.
44 N.B. Piekielek et al. / Ecological Informatics 30 (2015) 40–48
For diagnostics, we produced the area under the receiver operatingcharacteristic curve (AUC) that ranges from 0.5 to 1.0. An AUC of 0.5 in-dicates performance no better than random and 1.0 indicates perfectmodel prediction. AUCmeasures above approximately 0.7 are generallyconsidered to be good and above 0.9 are excellent (Franklin, 2009, al-though see Lobo et al., 2008 for a critique of AUC). AUC scores quantifythe discriminatory power of models however, additional important as-pects of SDM performance include accuracy (of probabilities of pres-ence), and generalizability. As such, we also examined SDMcalibration, and predictor importance. Calibration plots graph themodel predicted probability of presence versus the observed prevalencefrom cross-validation of testing points (see Vaughan and Ormerod,2005 for a more complete description of calibration plots).We reportedthe slope and intercept of calibration plots where an intercept of ‘0’ andslope of ‘1’ indicate perfect model accuracy. Variable importance scoreswere quantified as one minus the correlation between the original pre-dictions and predictions produced with a single variable permuted sothat low correlations indicate that the permuted variable was impor-tant. In addition to aidingmodel interpretation, lower predictor variableimportance in testing versus training data can indicate that a particularpredictor–response relationship does not generalize well to other con-ditions. Thefinal step inmodel construction appliedfinalmodels to con-temporary environmental conditions to produce a spatially continuousdistribution map for each species.
2.7. Projecting future probability of occurrence
We used SDMs to project future probabilities of occurrence for eachspecies under eighteen climate change scenarios (nine GCMs and twoRCPs). Future projections were performed for three 30-year periods re-ferred to by their ending year (e.g., 2040, 2070, and 2099). The resultingcurrent and future species distributions were referred to in three ways:1) core habitat, areas that were always projected to be within the spe-cies range; 2) deteriorating habitat, areas that were projected to changefrom being within the species range to outside of its range in any of thethree future time periods; and 3) expanding habitat, areas that wereprojected to be outside of a species range under current conditionsand to be within a species range in future conditions. To highlight
Table 3Diagnostics of forest and shrub species distribution models in the Greater Yellowstone Ecosysplots.
Species Code Number of pr
Sagebrush communityArtemesia tridentataArtemesia tridentata spp. VaseyanaArtemesia tridentata Rydb. ssp. TripartiteArtemesia tridentata spp. WyomingensisArtemesia tridentata spp. tridentata
artr 251(0.10)
Lower treelineJuniper communityJuniperus scopulorum Juniperus occidentalis Juniperus osteosperma
jusc 198(0.08)
Limber pine pifl 266(0.11)
MontaneAspen potr 417
(0.17)Douglas fir psme 863
(0.35)Lodgepole pine pico 1190
(0.49)
Upper treelineEngelmann spruce pien 962
(0.39)Subalpine fir abla 533
(0.21)
areas of agreement among GCMs and delineate where active manage-ment is most feasible based on current management jurisdiction andpolicy, we used a GIS to produce summary statistics (area, mean eleva-tion of species distribution, and management jurisdiction) where themajority (at least five out of nine possible) GCMs predicted species oc-currence. This was done for each RCP separately. To examine the effectof focusing only on locations where the majority of GCM results agree,we also reported the standard deviation of summary statistics basedon different levels of GCM agreement (varied one through nine possi-ble) divided by the total predicted currently suitable area.
3. Results
3.1. Model results
The discriminatory power of final models as measured by AUC wasgood (N0.7) or excellent (N0.9) for all species except for limber pine(0.655) (Table 3). The accuracy of models as measured by the slopeand intercept of calibration plots was good (slopeswithin 0.1 of 1, inter-cepts within 0.1 of 0) for all models except for sagebrush (slope =0.855, intercept = −0.214) and limber pine (slope = 0.755,intercept = −0.487). Response curves were increasing, decreasing orunimodal (Table 4). Average August soil moisture deficit (deft9), Aprilsnowpack (pack4) and the fraction of sand in surface soils (sandfract)were among the most important predictor variables across all speciesSDMs.
3.2. Projected responses to climate change
When models were used to project future distributions under cli-mate change scenarios lower elevation species expanded their totalarea while montane and upper treeline species were projected to havedeteriorating habitat area. For every species in the study, expandinghabitat was projected to be on average in higher elevation settingsthan contemporary habitat which led to projected increases in area onfederal restricted lands and decreases in area on federal general lands(Table 5).
tem. Model accuracy here is represented by the slope (α) and intercept (β) of calibration
esent observations (proportion prevalence) Model discrimination Model accuracy
AUC Calibration
0.731 α = −0.214β = 0.855
0.961 α = −0.0164β = 1.03
0.655 α = −0.487β = 0.755
0.863 α = −0.0208β = 0.97
0.777 α = −0.0343β = 0.942
0.768 α = −0.00228β = 0.962
0.765 α = −0.0164β = 0.94
0.857 α = −0.0147β = 0.979
Table 4Predictor variables used in multivariate adaptive regression spline models and the shapeof their relationship to probability of presence through the generation of response curves.Superscripts show the rank order of variable importance (fromhigh to low) inmodel test-ing against a withheld portion of the dataset. Increasing and decreasing responses repre-sented approximately linear positive and negative relationships, and unimodalresponses were increasing and then decreasing. See Table 1 for predictor abbreviations.
Species Predictors by shape of relationship
Increasing Decreasing Unimodal
Sagebrushcommunity
deft91; srad3;rckvol4; soilm66
pack42 sandfract5
Lower treelineJunipercommunity
pack41 ; twi2
Limber pine sandfract4 pack45; srad6 rckvol1; deft92; soilm63
MontaneAspen rckvol2 deft91; pack43;
sandfract4; soilm65
Douglas fir twi4 srad5; soilm66;sandfract2; rckvol7
pack41; deft93
Lodgepolepine
soilm62 sandfract1; deft93;pack44; srad5; rckvol6
Upper treelineEngelmannspruce
pack44; srad5;twi6
deft91; soilm67 rckvol2; sandfract3
Subalpine fir srad3 sandfract1; deft96 pack42; soilm64; rckvol5
45N.B. Piekielek et al. / Ecological Informatics 30 (2015) 40–48
Upper treeline species were projected to experience the most dra-matic deterioration of habitat area (mean 85% decrease, range 80%–90%), when calculated as present day compared to RCP 8.5 for the2099 period (Figs. 2 and 3) (all summaries that follow are for thesame comparison). Montane species also showed substantial projecteddeterioration in habitat area with an average decrease of 73% (range60%–85%). Lower treeline species exhibited a mix of projected re-sponses with limber pine projecting a 29% deterioration and juniper aprojected expansion (55%). Sagebrush was projected to have a 40% ex-pansion in habitat area although note some small areas of deteriorationin contrast to juniper for which there was no projected deterioration(Fig. 2). Projected change in themean elevation of distributions ofmon-tane species moved up an average of 413 m (but lodgepole pine only298 m), upper treeline species an average of 375 m, lower treeline spe-cies an average of 269 m, and sagebrush 200 m.
The two emissions scenarios agreed on likely future outcomes al-though using projections based on RCP 8.5 resulted in more rapid anddramatic reductions in projected habitat area for deteriorating habitatspecies and somewhat larger expansion of distribution area forexpanding species.
3.3. Implications for management actions
The proportion of species distributions on federal restricted landswas projected to increase for all species and climate change scenarios(Table 5). Upper treeline species distributions in particular, wereprojected to move up in elevation from federal general to federal re-stricted lands. The proportion of species distributions on general federallands was projected to increase for lower elevation species both fromdistribution expansion (e.g., sagebrush, juniper) and projected changesinmean elevation (e.g., limber pine, aspen and Douglas fir). By the 2099time period under RCP 8.5, more than half of the projected sagebrushand juniper distributionswere on federal lands (combining federal gen-eral and federal restricted), compared to 42% and 33% at present.
4. Discussion
Projected decreases in spring and summer snowpack along with in-creasing late season soilmoisture deficit over the course of the next cen-tury (Chang et al., 2014) should result in a longer and drier growing
season than present and general deterioration of forest habitat in theGYE. This is consistent with the results of other studies that associatedexpected climate changes with projected increases in wildlife activity(Westerling et al., 2011) and forest pest outbreaks (Macfarlane et al.,2012). Some tree species may be able to track their preferred climateto higher elevations (assuming no dispersal or other limitations), al-though this does not guarantee that soil conditions at higher elevationswill be amenable to forest establishment.
As projected distributions of focal species migrated up in elevationthey became better represented on federal restricted lands where man-agement options are currently limited by agency policy. Uncertaintiesin modeling results along with continually changing research objectivesand natural resourcemanagement staffsmake it imperative that the rela-tionships between predictors and response used to build ecologicalmodels are easily understood and that models are easily constructedandmodified (i.e., reproducible) in an adaptivemanagement framework.Scientific workflow software like VisTrails:SAHM show great potential toalleviate some of the technical challenges ofmodeling climate change im-pacts to support natural resource management decision-making.
The results of the present study refined our understanding of the en-vironmental drivers of vegetation distributions in the GYE by identifyingthe seasonal climate and soil conditions that structure communitiesacross four elevation zones. Results suggested that upper treeline is theresult of lengthy periods of snow cover that limit forest species establish-ment and lower treeline is a response to seasonal water deficit. Juniperpresents an exception to this pattern in that its ability to draw fromdeep soil water stores combined with post European settlement fire ex-clusion has resulted in juniper invasion of sagebrush communities (i.e.,below treeline) throughout much of the Western Cordillera, includingthe GYE (Leffler and Caldwell, 2005; Lyford et al., 2003). Contrary toprior studies that show future climate of the Yellowstone plateau beingmost suitable for present day Great Basin shrub/scrub communities(e.g., Rehfeldt et al., 2012), the present results suggested that future cli-mate may benefit juniper communities, potentially at the expense ofsagebrush (Davies et al., 2014). This difference was the result of a focuson species and genotypes that are currently present in the GYE (GreatBasin shrub/scrub was not considered).
A principal uncertainty in modeling montane forest habitat con-cerned interactions between soil properties and climate. Under currentclimate conditions, Douglas fir, subalpine fir, and Engelmann spruce atthe lower end of their range are presently excluded from portions ofthe Yellowstone plateau by rhyolitic (i.e., sandy with poor nutrient con-tent) soils. It is not clear whether this is through a competitive interac-tion with lodgepole pine, a complex interaction with climate, orwhether they simply cannot grow on rhyolitic soils. Under the RCP 4.5scenario of only moderate climate change, the Douglas fir distributionwas projected to move upslope to occupy a core area across much ofthe Yellowstone plateau whereas under the RCP 8.5 scenario theprojected Douglas fir distribution covered only a small portion of theYellowstone plateau. This result was similar to that of Schrag et al.(2008) who also considered soil conditions and described opposingconclusions on the fate of upper treeline species depending onwhetherfuture precipitation increased or decreased. Under a future warmer andwetter climate, it may be that the rhyolitic soils of the Yellowstone pla-teau become suitable for Douglas fir, although this seems unlikely basedon current observations. The calibration of more local scalemodels and/or field experimentation (e.g., common garden experiments) wouldshed light on the likely future fate of Douglas fir on the Yellowstoneplateau.
Upper treeline species are almost certainly vulnerable to climatechange and the incorporation of an estimate of future snowpack in-creased our confidence in this result. Upper treeline species may besqueezed between competitively dominant species moving upslope(like Douglas fir if it can occupy the Yellowstone plateau) and eitherthe tops of mountains and/or unfavorable upslope conditions (Bellet al., 2014). Although the present study included rock volume as a
Table 5Area and location of projected suitable habitat by species and RCP scenario based onmajority agreement of nineGCMs. Area is presented in square kilometers for current and as percentagechange from current for projected future. Management zones are: 1 = private; 2 = private protected and nonfederal public; 3 = federal general; 4 = federal restricted; 5 = other. Ele-vation is inmeters. In parentheses following change in area percentages are standard deviations in area change when agreement among GCMs is varied from 1 (at least one GCMprojectsspecies occurrence), to nine (all GCMs considered have to project species occurrence).
Species Present 2040 2070 2099
Area (percent by managementzones 1, 2, 3, 4, 5)
Area (percent by management zones) Area (percent by management zones) Area (percent by management zones)
RCP 4.5Sagebrush community 132,252 (50, 2, 38, 4, 6) +17% (+/−10) (43, 2, 41, 8, 6) +23% (+/−15) (42, 2, 41, 9, 6) +31% (+/−16) (40, 2, 41, 12, 5)Elevation (range) 1795 (897–3230) 1879 (897–3608) 1905 (897–3608) 1940 (897–3608)
Lower treelineJuniper community 133,727 (58, 2, 31, 2, 7) +18% (+/−10) (53, 2, 35, 4, 6) +26% (+/−15) (50, 2, 37, 5, 6) +32% (+/−15) (48, 2, 38, 6, 6)
1684 (893–2849) 1757 (893–3195) 1790 (893–3195) 1815 (893–3246)Limber pine 104,874 (41, 2, 34, 17, 6) −13% (+/−12) (33, 3, 38, 22, 4) −8% (+/−19) (29, 2, 42, 22, 5) −22% (+/−20) (25, 2, 44, 24, 5)
2013 (917–4015) 2136 (917–4015) 2184 (917–4015) 2231 (1007–4015)
MontaneAspen 61,028 (38, 1, 50, 9, 2) −1% (+/−25) (24, 0, 52, 23, 1) −5% (+/−32) (19, 0, 49, 30, 2) −10% (+/−31) (15, 0, 45, 40, 0)
2091 (1048–3117) 2241 (1059–3512) 2301 (1135–3512) 2399 (1135–3772)Douglas fir 78,229 (34, 2, 47, 14, 3) −35% (+/−16) (15, 2, 55, 27, 1) −38% (+/−25) (11, 1, 55, 32, 1) −53% (+/−26) (6, 0, 52, 41, 1)
2086 (992–3833) 2283 (1099–3734) 2341 (1099–3577) 2429 (1110–3577)Lodgepole pine 54,199 (3, 0, 46, 49, 2) −28% (+/−24) (1, 0, 36, 62, 1) −42% (+/−36) (0, 0, 30, 69, 1) −50% (+/−38) (0, 0, 24, 75, 1)
2460 (1736–3833) 2560 (1811–3867) 2602 (1896–3842) 2631 (1964–3867)
Upper treelineEngelmann spruce 53,843 (1, 0, 30, 66, 3) −46% (+/−24) (0, 0 , 22, 75, 3) −61% (+/−36) (0, 0, 18, 79, 3) −77% (+/−38) (0, 0, 16, 81, 3)
2712 (1123–4015) 2864 (1123–4015) 2934 (1123–4015) 3021 (1123–4015)Subalpine fir 42,144 (0, 0, 24, 72, 4) −43% (+/−30) (0, 0, 19, 75, 6) −56% (+/−46) (0, 0, 18, 76, 6) −68% (+/−49) (0, 0, 16, 76, 8)
2797 (1368–4015) 2929 (1354–4015) 2982 (1354–4015) 3038 (1354–4015)
RCP 8.5Sagebrush community 132,252 (50, 2, 38, 4, 6) +18% (+/−9) (43, 2, 41, 7, 7) +28% (+/−16) (40, 2, 41, 11, 6) +40% (+/−17) (37, 1, 40, 16, 6)Elevation (range) 1795 (897–3230) 1878 (897–3608) 1929 (897–3711) 1995 (897–3771)
Lower treelineJuniper community 133,727 (58, 2, 31, 2, 7) +16% (+/−9) (54, 2, 34, 4, 6) +32% (+/−16) (48, 2, 38, 6, 6) +55% (+/−16) (41, 1, 39, 14, 5)
1684 (893–2849) 1749 (893–3195) 1813 (893–3246) 1928 (893–3608)Limber pine 104,874 (41, 2, 34, 17, 6) −15% (+/−12) (32, 3, 38, 22, 5) −37% (+/−21) (29, 3, 43, 21, 4) −29% (+/−21) (17, 2, 49, 28, 4)
2013 (917–4015) 2147 (917–4015) 2189 (971–4015) 2307 (1071–4015)
MontaneAspen 61,028 (38, 1, 50, 9, 2) +7% (+/−25) (24, 0, 52, 22, 2) −1% (+/−23) (15, 0, 46, 38, 1) −60% (+/−36) (12, 0, 27, 61, 0)
2091 (1048–3117) 2234 (1059–3512) 2382 (1135–3772) 2560 (1356–3772)Douglas fir 78,229 (34, 2, 47, 14, 3) −37% (+/−16) (15, 2, 55, 27, 1) −63% (+/−27) (8, 0, 53, 37, 2) −73% (+/−28) (2, 0, 43, 53, 2)
2086 (992–3833) 2284 (1099–3577) 2394 (1110–3577) 2559 (1121–3714)Lodgepole pine 54,199 (3, 0, 46, 49, 2) −26% (+/−23) (1, 0, 36, 62, 0) −53% (+/−40) (0, 0, 24, 76, 0) −85% (+/−41) (0, 0, 11, 89, 0)
2460 (1736–3833) 2550 (1811–3867) 2622 (1964–3842) 2758 (2130–3833)
Upper treelineEngelmann spruce 53,843 (1, 0, 30, 66, 3) −47% (+/−23) (0, 0, 22, 76, 2) −77% (+/−40) (1, 0, 16, 80, 3) −90% (+/−41) (1, 0, 12, 84, 3)
2712 (1123–4015) 2864 (1123–4015) 3016 (1123–4015) 3145 (1123–4015)Subalpine fir 42,144 (0, 0, 24, 72, 4) −44% (+/−29) (0, 0, 20, 75, 5) −66% (+/−51) (0, 0, 16, 77, 7) −80% (+/−52) (0, 0, 12, 80, 8)
2797 (1368–4015) 2930 (1354–4015) 3036 (1354–4015) 3114 (1394–4015)
46 N.B. Piekielek et al. / Ecological Informatics 30 (2015) 40–48
predictor, upper treeline species models projected future distributionsup to the highest elevations in the study area where a lack of soil inmany places (i.e., exposed rock outcroppings) does not presently sup-port forest establishment. Upper treeline models missed an upper ele-vation habitat threshold due to a paucity of forest inventoryobservations above treeline and therefore a more weak response tohigh rock volume than we understand. There is evidence that uppertreeline species can rapidly invade alpine habitats where there are suit-able soil conditions and when climate is conducive, such as the climatethat is expected in the future of the American mountain west (Grant,2012). Although an expansion of our study area to include the northernrange limits of upper treeline species may have alleviated this issue byincluding upper elevation climatic limits to forest distributions, moreobservations of forest dynamics at and above treeline remains sorelyneeded to better understand upper treeline dynamics within the con-text of climate change impacts.
Both poor model performance and disagreement among GCM pro-jections limited our ability to draw conclusions from some SDM results.The limber pine SDM in particular did not performwell (low AUC, poor
model calibration), perhaps in part due to low species prevalence. Pro-jections of change in distribution area varied widely for a few specieswhen the level of agreement between GCMs was changed from lowlevels to high levels of agreement required to project future occurrence.This led to some area change projections whose uncertainty (standarddeviation of projected area change when agreement was varied),crossed zero (e.g., the results for aspen and limber pine). There wasalso substantial variation in the lodgepole pine results and to a lesser ex-tent the results for subalpine fir. However, the aspen, limber pine andlodgepole models were the only three for which projections of distribu-tion area change actually changed sign (i.e., deteriorating to expandingor vice versa) when levels of GCM agreement were varied while otherspecies models merely varied (sometimes greatly), in the extent towhich they projected expansion or deterioration of distributions. Exam-inations of uncertainty such as the one employed here highlight thelevels of disagreement among current GCM projections and the waysin which they can decrease our confidence in SDM results.
Modeling the potential ecological impacts of climate change to sup-port natural resource management decision-making presents a host of
Fig. 2. Current and projected future (RCP 8.5, majority agreement among GCMs) suitable habitat for eight vegetation species across four elevation zones of the Greater Yellowstone Eco-system. Species abbreviations are presented in Table 3. The black outline shows the National Park Service boundary [2-column fitting image, color reproduction online only].
47N.B. Piekielek et al. / Ecological Informatics 30 (2015) 40–48
technical challenges duringmodel formulation, evaluation, and technol-ogy transfer phases of a project. Many of these challenges are alleviatedby the use of custom scientific workflow software like VisTrails:SAHMthat enables the relatively rapid production of ecologically defensibleand well-documented SDMs. Time saved by custom software on themodel formulation phase of a project can be used tomore critically eval-uate future projection results, such as by zones of management jurisdic-tion in aGIS. The data andworkflowprovenance automatically capturedby scientific workflow software should also aid in the technology trans-fer of SDM results, such as to natural resource management agenciesthat want to run models in their own computing environment, updateresults with new or different GCMs, or develop and test their own eco-logical hypotheses in an adaptivemanagement framework. By engagingand empowering natural resource managers to also be producers ofclimate change understanding after the end of a sponsored research
project, they will likely be in a better position to develop successful cli-mate adaptation strategies.
Responding to expected future climate change is a daunting chal-lenge faced by today's natural resourcemanagement community. Fortu-nately, there is reason to be hopeful that employing a mix of currentlysuccessful strategies and new approaches may produce desirable out-comes for the GYE. For example, active management that is well-coordinated across ownership jurisdictions like the whitebark pinestrategy (GYCC, 2011) employed by the Greater Yellowstone Coordinat-ing Committee is already making contributions to the maintenance ofan important upper treeline species within the study area (GYCC,2011). New guidance on how to conduct climate vulnerability assess-ments (Glick et al., 2011) and how to respond to the results of those as-sessments (i.e., the Climate-smart framework for conservationplanning) (Stein et al., 2014) provide excellent guidance on hownatural
48 N.B. Piekielek et al. / Ecological Informatics 30 (2015) 40–48
resource managers can begin to respond to expected future climatechange.
Acknowledgments
Wewould like to thank theUSGSNorth Central Climate Science Cen-ter and Jeff Morisette for funding and the creation of VisTrails:SAHM.Funding was also provided by the NASA Applied Sciences Program(10-BIOCLIM10-0034) and the Montana NSF EPSCoR Initiative.
We would also like to thank Colin and Marian Talbert for their workon SAHMand their invaluable assistance in its use.We thank Linda Phil-lips for assistance creating graphics and D. Schlaepfer, P. Jantz and twoanonymous reviewers for insightful commentson a draft of thismanuscript.
We acknowledge the World Climate Research Programme's Work-ing Group on Coupled Modelling, which is responsible for CMIP, andwe thank the climate modeling groups (listed in Appendix S1, Table 3of this paper) for producing and making available their model output.For CMIP the U.S. Department of Energy's Program for Climate ModelDiagnosis and Intercomparison provides coordinating support and leddevelopment of software infrastructure in partnership with the GlobalOrganization for Earth System Science Portals.
References
Anderson, R.P., 2013. A framework for using niche models to estimate impacts of climatechange on species distributions. Ann. N. Y. Acad. Sci. 1297, 8–28. http://dx.doi.org/10.1111/nyas.12264.
Araújo, M.B., Guisan, A., 2006. Five (or so) challenges for species distribution modelling.J. Biogeogr. 33, 1677–1688. http://dx.doi.org/10.1111/j.1365-2699.2006.01584.x.
Bell, D.M., Bradford, J.B., Lauenroth, W.K., 2014. Mountain landscapes offer few opportu-nities for high‐elevation tree species migration. Glob. Chang. Biol. 20, 1441–1451.http://dx.doi.org/10.1111/gcb.12504.
Brown, K., Hansen, A.J., Keane, R.E., Graumlich, L.J., 2006. Complex interactions shapingaspen dynamics in the Greater Yellowstone Ecosystem. Landsc. Ecol. 21, 933–951.http://dx.doi.org/10.1007/s10980-005-6190-3.
Chang, T., Hansen, A.J., Piekielek, N., 2014. Patterns and variability of projected bioclimatichabitat for Pinus albicaulis in the Greater Yellowstone Area. PLoS ONE 9 (11),e111669.
Coops, N.C., Waring, R.H., 2010. A process-based approach to estimate lodgepole pine(Pinus contorta Dougl.) distribution in the Pacific Northwest under climate change.Clim. Chang. 105, 313–328. http://dx.doi.org/10.1007/s10584-010-9861-2.
Daly, C., Halbleib, M., Smith, J.I., Gibson, W.P., Doggett, M.K., Taylor, G.H., Pasteris, P.P.,2008. Physiographically sensitive mapping of climatological temperature and precip-itation across the conterminous united states. Int. J. Climatol. 28 (15), 2031–2064.http://dx.doi.org/10.1002/joc.1688.
Davies, K.W., Bates, J.D., Madsen, M.D., Nafus, A.M., 2014. Restoration of mountain bigsagebrush steppe following prescribed burning to control western juniper. Environ.Manage. 53, 1015–1022. http://dx.doi.org/10.1007/s00267-014-0255-5.
Despain, Don G., 1990. Yellowstone Vegetation: Consequences of Environment and Histo-ry in a Natural Setting. Roberts Rinehart Publishers, Inc.
ESRI, 2009. ArcGIS Desktop: Release 9.3. Environmental Systems Research Institute, Red-lands, CA.
Freire, J., Silva, C.T., 2012. Making computations and publications reproducible withVisTrails. Comput. Sci. Eng. http://dx.doi.org/10.1109/MCSE.2012.76 (July/August).
Franklin, J., 2009. Mapping Species Distributions: Spatial Inference and Prediction. Cam-bridge University Press, Cambridge, UK.
Gesch, D., Evans, G., Mauck, J., Hutchinson, J., Carswell Jr., W.J., 2009. The nationalmap—elevation. U.S. Geological Survey Fact Sheet 2009–3053 (4 pp.).
Gibson, J., Moisen, G., Frescino, T., Edwards, T.C., 2013. Using publicly available forest in-ventory data in climate-based models of tree species distribution: examining effectsof true versus altered location coordinates. Ecosystems 17, 43–53. http://dx.doi.org/10.1007/s10021-013-9703-y.
Glick, P., Stein, B.A., Edelson, N.A. (Eds.), 2011. Scanning the Conservation Horizon: AGuide to Climate Change Vulnerability Assessment. National Wildlife Federation,Washington, D.C.
Grant, E., 2012. Extrinsic regime shifts drive abrupt changes in regeneration dynamics atupper treeline in the Rocky Mountains, USA. Ecol. 93, 1614–1625. http://dx.doi.org/10.1890/11-1220.1.
Greater Yellowstone Coordinating Committee (GYCC), 2011. Whitebark Pine Subcommit-tee. Whitebark Pine Strategy for the Greater Yellowstone Area (41 pp.).
Hansen, A.J., Piekielek, N.B., Davis, C., Haas, J., Oliff, S.T., Gross, J., Monahan, B., Theobald, D.,2014. Exposure of U. S. National Parks to land use and climate change 1900–2100.Ecol. Appl. 24, 484–502. http://dx.doi.org/10.1890/13-0905.1.
Harte, J., Saleska, S., Shih, T., 2006. Shifts in plant dominance control carbon-cycle re-sponses to experimental warming and widespread drought. Environ. Res. Lett. 1,014001. http://dx.doi.org/10.1088/1748-9326/1/1/014001.
Knutti, R., Sedláček, J., 2012. Robustness and uncertainties in the new CMIP5 climate modelprojections. Nat. Clim. Chang. 3, 369–373. http://dx.doi.org/10.1038/NCLIMATE1716.
Leathwick, J.R., Elith, J., Hastie, T., 2006. Comparative performance of generalized additivemodels and multivariate adaptive regression splines for statistical modelling of speciesdistributions. Ecol. Model. 199, 188–196. http://dx.doi.org/10.1016/j.ecolmodel.2006.05.022.
Leffler, A.J., Caldwell, M.M., 2005. Shifts in depth of water extraction and photosyntheticcapacity inferred from stable isotope proxies across an ecotone of Juniperusosteosperma (Utah juniper) and Artemisia tridentata (big sagebrush). J. Ecol. 93,783–793. http://dx.doi.org/10.1111/j.1365-2745.2005.01014.x.
Littell, J.S., Peterson, D.L., Tjoelker, M., 2008. Douglas-fir growth in mountain ecosystem:water limits tree growth from stand to region. Ecol. Monogr. 78, 349–368. http://dx.doi.org/10.1890/07-0712.1.
Lobo, J.M., Jiménez-Valverde, A., Real, R., 2008. AUC: a misleading measure of the perfor-mance of predictive distribution models. Glob. Ecol. Biogeogr. 17, 145–151. http://dx.doi.org/10.1111/j.1466-8238.2007.00358.x.
Lutz, J.A., van Wagtendonk, J.W., Franklin, J.F., 2010. Climatic water deficit, tree speciesranges, and climate change in Yosemite National Park. J. Biogeogr. 37, 936–950.http://dx.doi.org/10.1111/j.1365-2699.2009.02268.x.
Lyford,M.E., Jackson, S.T., Betancourt, J.L., Gray, S.T., 2003. Influence of landscape structureand climate variability on a late Holocene plantmigration. Ecol. Monogr. 73, 567–583.http://dx.doi.org/10.1890/03-4011.
Macfarlane, W.W., Logan, J.A., Kern, W., 2012. An innovative aerial assessment of GreaterYellowstone Ecosystemmountain pine beetle-caused whitebark pine mortality. Ecol.Appl. 23 (2), 421–437.
McLane, S.C., Daniels, L.D., Aitken, S.N., 2011. Climate impacts on lodgepole pine (Pinuscontorta) radial growth in a provenance experiment. For. Ecol. Manag. 262, 115–123.http://dx.doi.org/10.1016/j.foreco.2011.03.007.
Means, R.E., 2011. Synthesis of lower treeline limber pine (Pinus flexis) woodland knowl-edge, research needs, and management considerations. The future of high-elevation,five-needle white pines in Western North America. In: Keane, R.E., Tomback, D.F.,Murray, M.P., Smith, C.M. (Eds.), Proceedings of the High Five Symposium, p. 376(Missoula, MT).
Miller, D.A., White, R.A., 1998. A conterminous United States multi-layer soil characteris-tics data set for regional climate and hydrology modeling. Earth Interact. 2.
Morisette, J.T., Jarnevich, C.S., Holcombe, T.R., Talbert, C.B., Ignizio, D., Talbert, M.K., Silva, C.,Koop, D., Swanson, A., Young, N.E., 2013. VisTrails:SAHM: visualization and workflowmanagement for species habitat modeling. Ecography 36, 129–135. http://dx.doi.org/10.1111/j.1600-0587.2012.07815.x.
Moss, R.H., Edmonds, J.A., Hibbard, K.A., Manning, M.R., Rose, S.K., Van Vuuren, D.P., et al.,2010. The next generation of scenarios for climate change research and assessment.Nature 463, 747–756. http://dx.doi.org/10.1038/nature08823.
O'Connell, B.M., LaPoint, E.B., Turner, J.A., Ridley, T., Boyer, D., Wilson, A.M., Waddell, K.L.,Christensen, G., Conkling, B.L., 2012. The Forest Inventory and Analysis Database: Da-tabase Description and Users Manual Version 5.1.2 for Phase 2. U.S. Department ofAgriculture.
Piekielek, N.B., Hansen, A.J., 2012. Extent of fragmentation of coarse-scale habitats in andaround U.S. National Parks. Biol. Conserv. 155, 13–22.
Rehfeldt, G.E., Crookston, N.L., Saenz-Romero, C., Cambell, E.M., 2012. North Americanvegetation model for land-use planning in a changing climate: a solution to largeclassification problems. Ecol. Appl. 22, 119–141. http://dx.doi.org/10.1890/11-0495.1.
Rupp, D.E., Abatzoglou, J.T., Hegewisch, K.C., Mote, P.W., 2013. Evaluation of CMIP5 20thcentury climate simulations for the Pacific Northwest USA. J. Geophys. Res. Atmos.118, 10,884–10,906. http://dx.doi.org/10.1002/jgrd.50843.
Schlaepfer, D.R., Lauenroth,W.K., Bradford, J.B., 2012. Consequences of declining snow ac-cumulation for water balance of mid-latitude dry regions. Glob. Chang. Biol. 18,1988–1997. http://dx.doi.org/10.1111/j.1365-2486.2012.02642.x.
Schrag, A.M., Bunn, A.G., Graumlich, L.J., 2008. Influence of bioclimatic variables on tree-line conifer distribution in the Greater Yellowstone Ecosystem: implications for spe-cies of conservation concern. J. Biogeogr. 35, 698–710. http://dx.doi.org/10.1111/j.1365-2699.2007.01815.x.
Stein, B.A., Glick, P., Edelson, N., Staudt, A. (Eds.), 2014. Climate-smart Conservation: Put-ting Adaptation Principles Into Practice. National Wildlife Federation, WashingtonD.C.
Stephenson, N.L., 1998. Actual evapotranspiration and deficit: biologically meaningfulcorrelates of vegetation distributions across spatial scales. J. Biogeogr. 25, 855–870.http://dx.doi.org/10.1046/j.1365-2699.1998.00233.x.
Thrasher, B., Xiong, J., Wang, W., Melton, F., Michaelis, A., Nemani, R., 2013. Downscaledclimate projections suitable for resource management. EOS Trans. Am. Geophys.Union 94, 321–323. http://dx.doi.org/10.1002/2013EO370002.
Vaughan, I.P., Ormerod, S.J., 2005. The continuing challenges of testing species distribu-tion models. J. Appl. Ecol. 42, 720–730. http://dx.doi.org/10.1111/j.1365-2664.2005.01052.x.
Westerling, A.L., Turner, M.G., Smithwick, E.A., Romme, W.H., Ryan, M.G., 2011. Continuedwarming could transform Greater Yellowstone fire regimes by mid-21st century.Proc. Natl. Acad. Sci. U. S. A. 108, 13165–13170.
Wilson, B.T., Lister, A.J., Riemann, R.I., 2012. A nearest-neighbor imputation approach tomapping tree species over large areas using forest inventory plots and moderate res-olution raster data. For. Ecol. Manag. 271, 182–198.
